Feature Engineering

An important step for successful machine learning is to choose what is the input of the algorithm. Meaning which information does the model (e.g. a classifier) get in order to predict an outcome. Features are attributes, that all subjects share. For example for a color image the intensities of each channel (red, green and blue) of each pixel can be considered as features. Besides this maybe obvious information, meta-data like the time or the place where the image was captured can be used as a feature. Furthermore, derived data like the NDVI, which can be computed for each pixel, or dense depth information can be used as additional features. In most of the classical machine learning approaches the performance highly depends on a good feature selection. This process of designing and selecting proper features is called feature engineering.

Related

Online Resources